From: Michael G. <mg...@te...> - 2004-06-22 21:51:23
|
> >If you can provide a small selfcontained example I'd be willing > >to test this using gcc-g77-3.3.3-20040217-1.tar.gz > > > I've attached a small example code which I just made up and tried and am= =20 > experiencing somewhat similar time ratio differences but it illustrates=20 > the point. (The attached code does nothing but calculate a bunch of=20 > sin's, cos's, tan's, and cexp's). Similar to my original post, I=20 > provide the following times on a 733 MHz machine. >=20 > Case A : 27.9 (Linux ) > Case B : 63.3 (Cygwin g77 using -mno-cygwin) > Case C: 47.4 (Cygwin g77) > Case D: 30.2 (Digital Fortran) > Case E : 64.9 (Mingw32 g77) I've compiled your example with no special setting whatsoever (i.e. g77 spdtst.f -o spdtst) and have taken the times given by the resulting executable. Case A: 20.8 stddev 4.6 (Linux, g77 3.3.3) Case B: 55.7 stddev 2.5 (WXP, linuxbased MinGW g77 x-compiler 3.3.3) Case c: 62.1 stddev 2.7 (W2K under linuxbased VMware, linuxbased MinGW g77 = x-compiler 3.3.3) Case D: 46.5 stddev 7.7 (WXP, MinGW g77 3.3.3) Case E: 64.9 stddev 8.0 (W2K under linuxbased VMware, MinGW g77 3.3.3) I've run the tests several times. Numbers given are the means and their respective stddev. The native linux executable is 17177 bytes The native W32 executable is 68940 bytes The linuxbased x-compiled executable is 1605477 bytes I think I could severely reduce the size of the x-compiled one but haven't tried. I'm somewhat surprised by this result and have no idea what the reason might be. The example invokes 30000000 x sin, cos, tan and 15000000 x cexp, cmplx. If I understand the libf2c correctly, neither of these functions makes use of pow. I have no clue why the execution speed differs by such a large margin (about a factor of 2.5 on my machine with no optimization involved) Best, Michael =2D-=20 Vote against SPAM - see http://www.politik-digital.de/spam/ Michael Gerdau email: mg...@te... GPG-keys available on request or at public keyserver |